xinyuqu/llm-review

This repository, called llm-review, is a plugin for AI coding tools (today Claude Code and Codex) that sends your git diff to six other large language models so each one can review your code. The reviewers are Claude, Cursor, DeepSeek, Gemini 3.5 Flash, Kimi, and MiniMax. You can ask just one of them for a quick second opinion, or run the whole panel at once for a consensus verdict before merging. The pitch behind using many models is that a model reviewing its own work tends to share its own blind spots. A different vendor's model has different biases and may catch problems the original missed. Bugs flagged by two or more panelists count as high-confidence findings; bugs flagged by only one are still worth a look. Several panel members, like Gemini Flash, Kimi, DeepSeek, and MiniMax, cost less per token than flagship reviewers, so fanning a diff across the panel is presented as affordable rather than extravagant. Inside the AI coding tool, the user gets a set of slash commands. The main one is /llm-review:panel, which checks which providers are working and then runs them in parallel before producing a single combined summary. There are also per-model commands like /llm-review:kimi or /llm-review:gemini, a /llm-review:status check that shows which providers are wired up, guided per-provider setup commands, and a /llm-review:jobs command for inspecting long reviews that run in the background. Setup happens one provider at a time. For each provider, you can supply an API key or point the plugin at the provider's official command-line tool and reuse the subscription you already pay for. Requirements are modest: Node.js 18 or newer, git, and a plugin-capable AI coding tool. Every review ends with a list of findings tagged by severity and a final verdict of SHIP, SHIP WITH FIXES, or DO NOT SHIP. The project is at version 1.0.0.